sys/netinet/siftr.c

   1 /*-
   2  * Copyright (c) 2007-2009
   3  *      Swinburne University of Technology, Melbourne, Australia.
   4  * Copyright (c) 2009-2010, The FreeBSD Foundation
   5  * All rights reserved.
   6  *
   7  * Portions of this software were developed at the Centre for Advanced
   8  * Internet Architectures, Swinburne University of Technology, Melbourne,
   9  * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 /******************************************************
  34  * Statistical Information For TCP Research (SIFTR)
  35  *
  36  * A FreeBSD kernel module that adds very basic intrumentation to the
  37  * TCP stack, allowing internal stats to be recorded to a log file
  38  * for experimental, debugging and performance analysis purposes.
  39  *
  40  * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst
  41  * working on the NewTCP research project at Swinburne University of
  42  * Technology's Centre for Advanced Internet Architectures, Melbourne,
  43  * Australia, which was made possible in part by a grant from the Cisco
  44  * University Research Program Fund at Community Foundation Silicon Valley.
  45  * More details are available at:
  46  *   http://caia.swin.edu.au/urp/newtcp/
  47  *
  48  * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of
  49  * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009.
  50  * More details are available at:
  51  *   http://www.freebsdfoundation.org/
  52  *   http://caia.swin.edu.au/freebsd/etcp09/
  53  *
  54  * Lawrence Stewart is the current maintainer, and all contact regarding
  55  * SIFTR should be directed to him via email: lastewart@swin.edu.au
  56  *
  57  * Initial release date: June 2007
  58  * Most recent update: September 2010
  59  ******************************************************/
  60
  61 #include <sys/cdefs.h>
  62 __FBSDID("$FreeBSD$");
  63
  64 #include <sys/param.h>
  65 #include <sys/alq.h>
  66 #include <sys/errno.h>
  67 #include <sys/eventhandler.h>
  68 #include <sys/hash.h>
  69 #include <sys/kernel.h>
  70 #include <sys/kthread.h>
  71 #include <sys/lock.h>
  72 #include <sys/mbuf.h>
  73 #include <sys/module.h>
  74 #include <sys/mutex.h>
  75 #include <sys/pcpu.h>
  76 #include <sys/proc.h>
  77 #include <sys/sbuf.h>
  78 #include <sys/smp.h>
  79 #include <sys/socket.h>
  80 #include <sys/socketvar.h>
  81 #include <sys/sysctl.h>
  82 #include <sys/unistd.h>
  83
  84 #include <net/if.h>
  85 #include <net/if_var.h>
  86 #include <net/pfil.h>
  87
  88 #include <netinet/in.h>
  89 #include <netinet/in_pcb.h>
  90 #include <netinet/in_systm.h>
  91 #include <netinet/in_var.h>
  92 #include <netinet/ip.h>
  93 #include <netinet/tcp_var.h>
  94
  95 #ifdef SIFTR_IPV6
  96 #include <netinet/ip6.h>
  97 #include <netinet6/in6_pcb.h>
  98 #endif /* SIFTR_IPV6 */
  99
 100 #include <machine/in_cksum.h>
 101
 102 /*
 103  * Three digit version number refers to X.Y.Z where:
 104  * X is the major version number
 105  * Y is bumped to mark backwards incompatible changes
 106  * Z is bumped to mark backwards compatible changes
 107  */
 108 #define V_MAJOR         1
 109 #define V_BACKBREAK     2
 110 #define V_BACKCOMPAT    4
 111 #define MODVERSION      __CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT))
 112 #define MODVERSION_STR  __XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \
 113     __XSTRING(V_BACKCOMPAT)
 114
 115 #define HOOK 0
 116 #define UNHOOK 1
 117 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536
 118 #define SYS_NAME "FreeBSD"
 119 #define PACKET_TAG_SIFTR 100
 120 #define PACKET_COOKIE_SIFTR 21749576
 121 #define SIFTR_LOG_FILE_MODE 0644
 122 #define SIFTR_DISABLE 0
 123 #define SIFTR_ENABLE 1
 124
 125 /*
 126  * Hard upper limit on the length of log messages. Bump this up if you add new
 127  * data fields such that the line length could exceed the below value.
 128  */
 129 #define MAX_LOG_MSG_LEN 200
 130 /* XXX: Make this a sysctl tunable. */
 131 #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN)
 132
 133 /*
 134  * 1 byte for IP version
 135  * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes
 136  * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes
 137  */
 138 #ifdef SIFTR_IPV6
 139 #define FLOW_KEY_LEN 37
 140 #else
 141 #define FLOW_KEY_LEN 13
 142 #endif
 143
 144 #ifdef SIFTR_IPV6
 145 #define SIFTR_IPMODE 6
 146 #else
 147 #define SIFTR_IPMODE 4
 148 #endif
 149
 150 /* useful macros */
 151 #define CAST_PTR_INT(X) (*((int*)(X)))
 152
 153 #define UPPER_SHORT(X)  (((X) & 0xFFFF0000) >> 16)
 154 #define LOWER_SHORT(X)  ((X) & 0x0000FFFF)
 155
 156 #define FIRST_OCTET(X)  (((X) & 0xFF000000) >> 24)
 157 #define SECOND_OCTET(X) (((X) & 0x00FF0000) >> 16)
 158 #define THIRD_OCTET(X)  (((X) & 0x0000FF00) >> 8)
 159 #define FOURTH_OCTET(X) ((X) & 0x000000FF)
 160
 161 static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR");
 162 static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode",
 163     "SIFTR pkt_node struct");
 164 static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode",
 165     "SIFTR flow_hash_node struct");
 166
 167 /* Used as links in the pkt manager queue. */
 168 struct pkt_node {
 169         /* Timestamp of pkt as noted in the pfil hook. */
 170         struct timeval          tval;
 171         /* Direction pkt is travelling; either PFIL_IN or PFIL_OUT. */
 172         uint8_t                 direction;
 173         /* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */
 174         uint8_t                 ipver;
 175         /* Hash of the pkt which triggered the log message. */
 176         uint32_t                hash;
 177         /* Local/foreign IP address. */
 178 #ifdef SIFTR_IPV6
 179         uint32_t                ip_laddr[4];
 180         uint32_t                ip_faddr[4];
 181 #else
 182         uint8_t                 ip_laddr[4];
 183         uint8_t                 ip_faddr[4];
 184 #endif
 185         /* Local TCP port. */
 186         uint16_t                tcp_localport;
 187         /* Foreign TCP port. */
 188         uint16_t                tcp_foreignport;
 189         /* Congestion Window (bytes). */
 190         u_long                  snd_cwnd;
 191         /* Sending Window (bytes). */
 192         u_long                  snd_wnd;
 193         /* Receive Window (bytes). */
 194         u_long                  rcv_wnd;
 195         /* Unused (was: Bandwidth Controlled Window (bytes)). */
 196         u_long                  snd_bwnd;
 197         /* Slow Start Threshold (bytes). */
 198         u_long                  snd_ssthresh;
 199         /* Current state of the TCP FSM. */
 200         int                     conn_state;
 201         /* Max Segment Size (bytes). */
 202         u_int                   max_seg_size;
 203         /*
 204          * Smoothed RTT stored as found in the TCP control block
 205          * in units of (TCP_RTT_SCALE*hz).
 206          */
 207         int                     smoothed_rtt;
 208         /* Is SACK enabled? */
 209         u_char                  sack_enabled;
 210         /* Window scaling for snd window. */
 211         u_char                  snd_scale;
 212         /* Window scaling for recv window. */
 213         u_char                  rcv_scale;
 214         /* TCP control block flags. */
 215         u_int                   flags;
 216         /* Retransmit timeout length. */
 217         int                     rxt_length;
 218         /* Size of the TCP send buffer in bytes. */
 219         u_int                   snd_buf_hiwater;
 220         /* Current num bytes in the send socket buffer. */
 221         u_int                   snd_buf_cc;
 222         /* Size of the TCP receive buffer in bytes. */
 223         u_int                   rcv_buf_hiwater;
 224         /* Current num bytes in the receive socket buffer. */
 225         u_int                   rcv_buf_cc;
 226         /* Number of bytes inflight that we are waiting on ACKs for. */
 227         u_int                   sent_inflight_bytes;
 228         /* Number of segments currently in the reassembly queue. */
 229         int                     t_segqlen;
 230         /* Link to next pkt_node in the list. */
 231         STAILQ_ENTRY(pkt_node)  nodes;
 232 };
 233
 234 struct flow_hash_node
 235 {
 236         uint16_t counter;
 237         uint8_t key[FLOW_KEY_LEN];
 238         LIST_ENTRY(flow_hash_node) nodes;
 239 };
 240
 241 struct siftr_stats
 242 {
 243         /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */
 244         uint64_t n_in;
 245         uint64_t n_out;
 246         /* # pkts skipped due to failed malloc calls. */
 247         uint32_t nskip_in_malloc;
 248         uint32_t nskip_out_malloc;
 249         /* # pkts skipped due to failed mtx acquisition. */
 250         uint32_t nskip_in_mtx;
 251         uint32_t nskip_out_mtx;
 252         /* # pkts skipped due to failed inpcb lookups. */
 253         uint32_t nskip_in_inpcb;
 254         uint32_t nskip_out_inpcb;
 255         /* # pkts skipped due to failed tcpcb lookups. */
 256         uint32_t nskip_in_tcpcb;
 257         uint32_t nskip_out_tcpcb;
 258         /* # pkts skipped due to stack reinjection. */
 259         uint32_t nskip_in_dejavu;
 260         uint32_t nskip_out_dejavu;
 261 };
 262
 263 static DPCPU_DEFINE(struct siftr_stats, ss);
 264
 265 static volatile unsigned int siftr_exit_pkt_manager_thread = 0;
 266 static unsigned int siftr_enabled = 0;
 267 static unsigned int siftr_pkts_per_log = 1;
 268 static unsigned int siftr_generate_hashes = 0;
 269 /* static unsigned int siftr_binary_log = 0; */
 270 static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log";
 271 static u_long siftr_hashmask;
 272 STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue);
 273 LIST_HEAD(listhead, flow_hash_node) *counter_hash;
 274 static int wait_for_pkt;
 275 static struct alq *siftr_alq = NULL;
 276 static struct mtx siftr_pkt_queue_mtx;
 277 static struct mtx siftr_pkt_mgr_mtx;
 278 static struct thread *siftr_pkt_manager_thr = NULL;
 279 /*
 280  * pfil.h defines PFIL_IN as 1 and PFIL_OUT as 2,
 281  * which we use as an index into this array.
 282  */
 283 static char direction[3] = {'\0', 'i','o'};
 284
 285 /* Required function prototypes. */
 286 static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS);
 287 static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS);
 288
 289
 290 /* Declare the net.inet.siftr sysctl tree and populate it. */
 291
 292 SYSCTL_DECL(_net_inet_siftr);
 293
 294 SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL,
 295     "siftr related settings");
 296
 297 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW,
 298     &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU",
 299     "switch siftr module operations on/off");
 300
 301 SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW,
 302     &siftr_logfile, sizeof(siftr_logfile), &siftr_sysctl_logfile_name_handler,
 303     "A", "file to save siftr log messages to");
 304
 305 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW,
 306     &siftr_pkts_per_log, 1,
 307     "number of packets between generating a log message");
 308
 309 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW,
 310     &siftr_generate_hashes, 0,
 311     "enable packet hash generation");
 312
 313 /* XXX: TODO
 314 SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW,
 315     &siftr_binary_log, 0,
 316     "write log files in binary instead of ascii");
 317 */
 318
 319
 320 /* Begin functions. */
 321
 322 static void
 323 siftr_process_pkt(struct pkt_node * pkt_node)
 324 {
 325         struct flow_hash_node *hash_node;
 326         struct listhead *counter_list;
 327         struct siftr_stats *ss;
 328         struct ale *log_buf;
 329         uint8_t key[FLOW_KEY_LEN];
 330         uint8_t found_match, key_offset;
 331
 332         hash_node = NULL;
 333         ss = DPCPU_PTR(ss);
 334         found_match = 0;
 335         key_offset = 1;
 336
 337         /*
 338          * Create the key that will be used to create a hash index
 339          * into our hash table. Our key consists of:
 340          * ipversion, localip, localport, foreignip, foreignport
 341          */
 342         key[0] = pkt_node->ipver;
 343         memcpy(key + key_offset, &pkt_node->ip_laddr,
 344             sizeof(pkt_node->ip_laddr));
 345         key_offset += sizeof(pkt_node->ip_laddr);
 346         memcpy(key + key_offset, &pkt_node->tcp_localport,
 347             sizeof(pkt_node->tcp_localport));
 348         key_offset += sizeof(pkt_node->tcp_localport);
 349         memcpy(key + key_offset, &pkt_node->ip_faddr,
 350             sizeof(pkt_node->ip_faddr));
 351         key_offset += sizeof(pkt_node->ip_faddr);
 352         memcpy(key + key_offset, &pkt_node->tcp_foreignport,
 353             sizeof(pkt_node->tcp_foreignport));
 354
 355         counter_list = counter_hash +
 356             (hash32_buf(key, sizeof(key), 0) & siftr_hashmask);
 357
 358         /*
 359          * If the list is not empty i.e. the hash index has
 360          * been used by another flow previously.
 361          */
 362         if (LIST_FIRST(counter_list) != NULL) {
 363                 /*
 364                  * Loop through the hash nodes in the list.
 365                  * There should normally only be 1 hash node in the list,
 366                  * except if there have been collisions at the hash index
 367                  * computed by hash32_buf().
 368                  */
 369                 LIST_FOREACH(hash_node, counter_list, nodes) {
 370                         /*
 371                          * Check if the key for the pkt we are currently
 372                          * processing is the same as the key stored in the
 373                          * hash node we are currently processing.
 374                          * If they are the same, then we've found the
 375                          * hash node that stores the counter for the flow
 376                          * the pkt belongs to.
 377                          */
 378                         if (memcmp(hash_node->key, key, sizeof(key)) == 0) {
 379                                 found_match = 1;
 380                                 break;
 381                         }
 382                 }
 383         }
 384
 385         /* If this flow hash hasn't been seen before or we have a collision. */
 386         if (hash_node == NULL || !found_match) {
 387                 /* Create a new hash node to store the flow's counter. */
 388                 hash_node = malloc(sizeof(struct flow_hash_node),
 389                     M_SIFTR_HASHNODE, M_WAITOK);
 390
 391                 if (hash_node != NULL) {
 392                         /* Initialise our new hash node list entry. */
 393                         hash_node->counter = 0;
 394                         memcpy(hash_node->key, key, sizeof(key));
 395                         LIST_INSERT_HEAD(counter_list, hash_node, nodes);
 396                 } else {
 397                         /* Malloc failed. */
 398                         if (pkt_node->direction == PFIL_IN)
 399                                 ss->nskip_in_malloc++;
 400                         else
 401                                 ss->nskip_out_malloc++;
 402
 403                         return;
 404                 }
 405         } else if (siftr_pkts_per_log > 1) {
 406                 /*
 407                  * Taking the remainder of the counter divided
 408                  * by the current value of siftr_pkts_per_log
 409                  * and storing that in counter provides a neat
 410                  * way to modulate the frequency of log
 411                  * messages being written to the log file.
 412                  */
 413                 hash_node->counter = (hash_node->counter + 1) %
 414                     siftr_pkts_per_log;
 415
 416                 /*
 417                  * If we have not seen enough packets since the last time
 418                  * we wrote a log message for this connection, return.
 419                  */
 420                 if (hash_node->counter > 0)
 421                         return;
 422         }
 423
 424         log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK);
 425
 426         if (log_buf == NULL)
 427                 return; /* Should only happen if the ALQ is shutting down. */
 428
 429 #ifdef SIFTR_IPV6
 430         pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]);
 431         pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]);
 432
 433         if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */
 434                 pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]);
 435                 pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]);
 436                 pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]);
 437                 pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]);
 438                 pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]);
 439                 pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]);
 440
 441                 /* Construct an IPv6 log message. */
 442                 log_buf->ae_bytesused = snprintf(log_buf->ae_data,
 443                     MAX_LOG_MSG_LEN,
 444                     "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:"
 445                     "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,"
 446                     "%u,%d,%u,%u,%u,%u,%u,%u\n",
 447                     direction[pkt_node->direction],
 448                     pkt_node->hash,
 449                     pkt_node->tval.tv_sec,
 450                     pkt_node->tval.tv_usec,
 451                     UPPER_SHORT(pkt_node->ip_laddr[0]),
 452                     LOWER_SHORT(pkt_node->ip_laddr[0]),
 453                     UPPER_SHORT(pkt_node->ip_laddr[1]),
 454                     LOWER_SHORT(pkt_node->ip_laddr[1]),
 455                     UPPER_SHORT(pkt_node->ip_laddr[2]),
 456                     LOWER_SHORT(pkt_node->ip_laddr[2]),
 457                     UPPER_SHORT(pkt_node->ip_laddr[3]),
 458                     LOWER_SHORT(pkt_node->ip_laddr[3]),
 459                     ntohs(pkt_node->tcp_localport),
 460                     UPPER_SHORT(pkt_node->ip_faddr[0]),
 461                     LOWER_SHORT(pkt_node->ip_faddr[0]),
 462                     UPPER_SHORT(pkt_node->ip_faddr[1]),
 463                     LOWER_SHORT(pkt_node->ip_faddr[1]),
 464                     UPPER_SHORT(pkt_node->ip_faddr[2]),
 465                     LOWER_SHORT(pkt_node->ip_faddr[2]),
 466                     UPPER_SHORT(pkt_node->ip_faddr[3]),
 467                     LOWER_SHORT(pkt_node->ip_faddr[3]),
 468                     ntohs(pkt_node->tcp_foreignport),
 469                     pkt_node->snd_ssthresh,
 470                     pkt_node->snd_cwnd,
 471                     pkt_node->snd_bwnd,
 472                     pkt_node->snd_wnd,
 473                     pkt_node->rcv_wnd,
 474                     pkt_node->snd_scale,
 475                     pkt_node->rcv_scale,
 476                     pkt_node->conn_state,
 477                     pkt_node->max_seg_size,
 478                     pkt_node->smoothed_rtt,
 479                     pkt_node->sack_enabled,
 480                     pkt_node->flags,
 481                     pkt_node->rxt_length,
 482                     pkt_node->snd_buf_hiwater,
 483                     pkt_node->snd_buf_cc,
 484                     pkt_node->rcv_buf_hiwater,
 485                     pkt_node->rcv_buf_cc,
 486                     pkt_node->sent_inflight_bytes,
 487                     pkt_node->t_segqlen);
 488         } else { /* IPv4 packet */
 489                 pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]);
 490                 pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]);
 491                 pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]);
 492                 pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]);
 493                 pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]);
 494                 pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]);
 495                 pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]);
 496                 pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]);
 497 #endif /* SIFTR_IPV6 */
 498
 499                 /* Construct an IPv4 log message. */
 500                 log_buf->ae_bytesused = snprintf(log_buf->ae_data,
 501                     MAX_LOG_MSG_LEN,
 502                     "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld,"
 503                     "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u,%u\n",
 504                     direction[pkt_node->direction],
 505                     pkt_node->hash,
 506                     (intmax_t)pkt_node->tval.tv_sec,
 507                     pkt_node->tval.tv_usec,
 508                     pkt_node->ip_laddr[0],
 509                     pkt_node->ip_laddr[1],
 510                     pkt_node->ip_laddr[2],
 511                     pkt_node->ip_laddr[3],
 512                     ntohs(pkt_node->tcp_localport),
 513                     pkt_node->ip_faddr[0],
 514                     pkt_node->ip_faddr[1],
 515                     pkt_node->ip_faddr[2],
 516                     pkt_node->ip_faddr[3],
 517                     ntohs(pkt_node->tcp_foreignport),
 518                     pkt_node->snd_ssthresh,
 519                     pkt_node->snd_cwnd,
 520                     pkt_node->snd_bwnd,
 521                     pkt_node->snd_wnd,
 522                     pkt_node->rcv_wnd,
 523                     pkt_node->snd_scale,
 524                     pkt_node->rcv_scale,
 525                     pkt_node->conn_state,
 526                     pkt_node->max_seg_size,
 527                     pkt_node->smoothed_rtt,
 528                     pkt_node->sack_enabled,
 529                     pkt_node->flags,
 530                     pkt_node->rxt_length,
 531                     pkt_node->snd_buf_hiwater,
 532                     pkt_node->snd_buf_cc,
 533                     pkt_node->rcv_buf_hiwater,
 534                     pkt_node->rcv_buf_cc,
 535                     pkt_node->sent_inflight_bytes,
 536                     pkt_node->t_segqlen);
 537 #ifdef SIFTR_IPV6
 538         }
 539 #endif
 540
 541         alq_post_flags(siftr_alq, log_buf, 0);
 542 }
 543
 544
 545 static void
 546 siftr_pkt_manager_thread(void *arg)
 547 {
 548         STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue =
 549             STAILQ_HEAD_INITIALIZER(tmp_pkt_queue);
 550         struct pkt_node *pkt_node, *pkt_node_temp;
 551         uint8_t draining;
 552
 553         draining = 2;
 554
 555         mtx_lock(&siftr_pkt_mgr_mtx);
 556
 557         /* draining == 0 when queue has been flushed and it's safe to exit. */
 558         while (draining) {
 559                 /*
 560                  * Sleep until we are signalled to wake because thread has
 561                  * been told to exit or until 1 tick has passed.
 562                  */
 563                 mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait",
 564                     1);
 565
 566                 /* Gain exclusive access to the pkt_node queue. */
 567                 mtx_lock(&siftr_pkt_queue_mtx);
 568
 569                 /*
 570                  * Move pkt_queue to tmp_pkt_queue, which leaves
 571                  * pkt_queue empty and ready to receive more pkt_nodes.
 572                  */
 573                 STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue);
 574
 575                 /*
 576                  * We've finished making changes to the list. Unlock it
 577                  * so the pfil hooks can continue queuing pkt_nodes.
 578                  */
 579                 mtx_unlock(&siftr_pkt_queue_mtx);
 580
 581                 /*
 582                  * We can't hold a mutex whilst calling siftr_process_pkt
 583                  * because ALQ might sleep waiting for buffer space.
 584                  */
 585                 mtx_unlock(&siftr_pkt_mgr_mtx);
 586
 587                 /* Flush all pkt_nodes to the log file. */
 588                 STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes,
 589                     pkt_node_temp) {
 590                         siftr_process_pkt(pkt_node);
 591                         STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes);
 592                         free(pkt_node, M_SIFTR_PKTNODE);
 593                 }
 594
 595                 KASSERT(STAILQ_EMPTY(&tmp_pkt_queue),
 596                     ("SIFTR tmp_pkt_queue not empty after flush"));
 597
 598                 mtx_lock(&siftr_pkt_mgr_mtx);
 599
 600                 /*
 601                  * If siftr_exit_pkt_manager_thread gets set during the window
 602                  * where we are draining the tmp_pkt_queue above, there might
 603                  * still be pkts in pkt_queue that need to be drained.
 604                  * Allow one further iteration to occur after
 605                  * siftr_exit_pkt_manager_thread has been set to ensure
 606                  * pkt_queue is completely empty before we kill the thread.
 607                  *
 608                  * siftr_exit_pkt_manager_thread is set only after the pfil
 609                  * hooks have been removed, so only 1 extra iteration
 610                  * is needed to drain the queue.
 611                  */
 612                 if (siftr_exit_pkt_manager_thread)
 613                         draining--;
 614         }
 615
 616         mtx_unlock(&siftr_pkt_mgr_mtx);
 617
 618         /* Calls wakeup on this thread's struct thread ptr. */
 619         kthread_exit();
 620 }
 621
 622
 623 static uint32_t
 624 hash_pkt(struct mbuf *m, uint32_t offset)
 625 {
 626         uint32_t hash;
 627
 628         hash = 0;
 629
 630         while (m != NULL && offset > m->m_len) {
 631                 /*
 632                  * The IP packet payload does not start in this mbuf, so
 633                  * need to figure out which mbuf it starts in and what offset
 634                  * into the mbuf's data region the payload starts at.
 635                  */
 636                 offset -= m->m_len;
 637                 m = m->m_next;
 638         }
 639
 640         while (m != NULL) {
 641                 /* Ensure there is data in the mbuf */
 642                 if ((m->m_len - offset) > 0)
 643                         hash = hash32_buf(m->m_data + offset,
 644                             m->m_len - offset, hash);
 645
 646                 m = m->m_next;
 647                 offset = 0;
 648         }
 649
 650         return (hash);
 651 }
 652
 653
 654 /*
 655  * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that
 656  * it's a reinjected packet and return. If it doesn't, tag the mbuf and return.
 657  * Return value >0 means the caller should skip processing this mbuf.
 658  */
 659 static inline int
 660 siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss)
 661 {
 662         if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL)
 663             != NULL) {
 664                 if (dir == PFIL_IN)
 665                         ss->nskip_in_dejavu++;
 666                 else
 667                         ss->nskip_out_dejavu++;
 668
 669                 return (1);
 670         } else {
 671                 struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR,
 672                     PACKET_TAG_SIFTR, 0, M_NOWAIT);
 673                 if (tag == NULL) {
 674                         if (dir == PFIL_IN)
 675                                 ss->nskip_in_malloc++;
 676                         else
 677                                 ss->nskip_out_malloc++;
 678
 679                         return (1);
 680                 }
 681
 682                 m_tag_prepend(m, tag);
 683         }
 684
 685         return (0);
 686 }
 687
 688
 689 /*
 690  * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL
 691  * otherwise.
 692  */
 693 static inline struct inpcb *
 694 siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport,
 695     uint16_t dport, int dir, struct siftr_stats *ss)
 696 {
 697         struct inpcb *inp;
 698
 699         /* We need the tcbinfo lock. */
 700         INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
 701
 702         if (dir == PFIL_IN)
 703                 inp = (ipver == INP_IPV4 ?
 704                     in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst,
 705                     dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
 706                     :
 707 #ifdef SIFTR_IPV6
 708                     in6_pcblookup(&V_tcbinfo,
 709                     &((struct ip6_hdr *)ip)->ip6_src, sport,
 710                     &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB,
 711                     m->m_pkthdr.rcvif)
 712 #else
 713                     NULL
 714 #endif
 715                     );
 716
 717         else
 718                 inp = (ipver == INP_IPV4 ?
 719                     in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src,
 720                     sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif)
 721                     :
 722 #ifdef SIFTR_IPV6
 723                     in6_pcblookup(&V_tcbinfo,
 724                     &((struct ip6_hdr *)ip)->ip6_dst, dport,
 725                     &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB,
 726                     m->m_pkthdr.rcvif)
 727 #else
 728                     NULL
 729 #endif
 730                     );
 731
 732         /* If we can't find the inpcb, bail. */
 733         if (inp == NULL) {
 734                 if (dir == PFIL_IN)
 735                         ss->nskip_in_inpcb++;
 736                 else
 737                         ss->nskip_out_inpcb++;
 738         }
 739
 740         return (inp);
 741 }
 742
 743
 744 static inline void
 745 siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp,
 746     int ipver, int dir, int inp_locally_locked)
 747 {
 748 #ifdef SIFTR_IPV6
 749         if (ipver == INP_IPV4) {
 750                 pn->ip_laddr[3] = inp->inp_laddr.s_addr;
 751                 pn->ip_faddr[3] = inp->inp_faddr.s_addr;
 752 #else
 753                 *((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr;
 754                 *((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr;
 755 #endif
 756 #ifdef SIFTR_IPV6
 757         } else {
 758                 pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0];
 759                 pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1];
 760                 pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2];
 761                 pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3];
 762                 pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0];
 763                 pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1];
 764                 pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2];
 765                 pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3];
 766         }
 767 #endif
 768         pn->tcp_localport = inp->inp_lport;
 769         pn->tcp_foreignport = inp->inp_fport;
 770         pn->snd_cwnd = tp->snd_cwnd;
 771         pn->snd_wnd = tp->snd_wnd;
 772         pn->rcv_wnd = tp->rcv_wnd;
 773         pn->snd_bwnd = 0;               /* Unused, kept for compat. */
 774         pn->snd_ssthresh = tp->snd_ssthresh;
 775         pn->snd_scale = tp->snd_scale;
 776         pn->rcv_scale = tp->rcv_scale;
 777         pn->conn_state = tp->t_state;
 778         pn->max_seg_size = tp->t_maxseg;
 779         pn->smoothed_rtt = tp->t_srtt;
 780         pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0;
 781         pn->flags = tp->t_flags;
 782         pn->rxt_length = tp->t_rxtcur;
 783         pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat;
 784         pn->snd_buf_cc = inp->inp_socket->so_snd.sb_cc;
 785         pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat;
 786         pn->rcv_buf_cc = inp->inp_socket->so_rcv.sb_cc;
 787         pn->sent_inflight_bytes = tp->snd_max - tp->snd_una;
 788         pn->t_segqlen = tp->t_segqlen;
 789
 790         /* We've finished accessing the tcb so release the lock. */
 791         if (inp_locally_locked)
 792                 INP_RUNLOCK(inp);
 793
 794         pn->ipver = ipver;
 795         pn->direction = dir;
 796
 797         /*
 798          * Significantly more accurate than using getmicrotime(), but slower!
 799          * Gives true microsecond resolution at the expense of a hit to
 800          * maximum pps throughput processing when SIFTR is loaded and enabled.
 801          */
 802         microtime(&pn->tval);
 803 }
 804
 805
 806 /*
 807  * pfil hook that is called for each IPv4 packet making its way through the
 808  * stack in either direction.
 809  * The pfil subsystem holds a non-sleepable mutex somewhere when
 810  * calling our hook function, so we can't sleep at all.
 811  * It's very important to use the M_NOWAIT flag with all function calls
 812  * that support it so that they won't sleep, otherwise you get a panic.
 813  */
 814 static int
 815 siftr_chkpkt(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
 816     struct inpcb *inp)
 817 {
 818         struct pkt_node *pn;
 819         struct ip *ip;
 820         struct tcphdr *th;
 821         struct tcpcb *tp;
 822         struct siftr_stats *ss;
 823         unsigned int ip_hl;
 824         int inp_locally_locked;
 825
 826         inp_locally_locked = 0;
 827         ss = DPCPU_PTR(ss);
 828
 829         /*
 830          * m_pullup is not required here because ip_{input|output}
 831          * already do the heavy lifting for us.
 832          */
 833
 834         ip = mtod(*m, struct ip *);
 835
 836         /* Only continue processing if the packet is TCP. */
 837         if (ip->ip_p != IPPROTO_TCP)
 838                 goto ret;
 839
 840         /*
 841          * If a kernel subsystem reinjects packets into the stack, our pfil
 842          * hook will be called multiple times for the same packet.
 843          * Make sure we only process unique packets.
 844          */
 845         if (siftr_chkreinject(*m, dir, ss))
 846                 goto ret;
 847
 848         if (dir == PFIL_IN)
 849                 ss->n_in++;
 850         else
 851                 ss->n_out++;
 852
 853         /*
 854          * Create a tcphdr struct starting at the correct offset
 855          * in the IP packet. ip->ip_hl gives the ip header length
 856          * in 4-byte words, so multiply it to get the size in bytes.
 857          */
 858         ip_hl = (ip->ip_hl << 2);
 859         th = (struct tcphdr *)((caddr_t)ip + ip_hl);
 860
 861         /*
 862          * If the pfil hooks don't provide a pointer to the
 863          * inpcb, we need to find it ourselves and lock it.
 864          */
 865         if (!inp) {
 866                 /* Find the corresponding inpcb for this pkt. */
 867                 inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport,
 868                     th->th_dport, dir, ss);
 869
 870                 if (inp == NULL)
 871                         goto ret;
 872                 else
 873                         inp_locally_locked = 1;
 874         }
 875
 876         INP_LOCK_ASSERT(inp);
 877
 878         /* Find the TCP control block that corresponds with this packet */
 879         tp = intotcpcb(inp);
 880
 881         /*
 882          * If we can't find the TCP control block (happens occasionaly for a
 883          * packet sent during the shutdown phase of a TCP connection),
 884          * or we're in the timewait state, bail
 885          */
 886         if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
 887                 if (dir == PFIL_IN)
 888                         ss->nskip_in_tcpcb++;
 889                 else
 890                         ss->nskip_out_tcpcb++;
 891
 892                 goto inp_unlock;
 893         }
 894
 895         pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
 896
 897         if (pn == NULL) {
 898                 if (dir == PFIL_IN)
 899                         ss->nskip_in_malloc++;
 900                 else
 901                         ss->nskip_out_malloc++;
 902
 903                 goto inp_unlock;
 904         }
 905
 906         siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked);
 907
 908         if (siftr_generate_hashes) {
 909                 if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) {
 910                         /*
 911                          * For outbound packets, the TCP checksum isn't
 912                          * calculated yet. This is a problem for our packet
 913                          * hashing as the receiver will calc a different hash
 914                          * to ours if we don't include the correct TCP checksum
 915                          * in the bytes being hashed. To work around this
 916                          * problem, we manually calc the TCP checksum here in
 917                          * software. We unset the CSUM_TCP flag so the lower
 918                          * layers don't recalc it.
 919                          */
 920                         (*m)->m_pkthdr.csum_flags &= ~CSUM_TCP;
 921
 922                         /*
 923                          * Calculate the TCP checksum in software and assign
 924                          * to correct TCP header field, which will follow the
 925                          * packet mbuf down the stack. The trick here is that
 926                          * tcp_output() sets th->th_sum to the checksum of the
 927                          * pseudo header for us already. Because of the nature
 928                          * of the checksumming algorithm, we can sum over the
 929                          * entire IP payload (i.e. TCP header and data), which
 930                          * will include the already calculated pseduo header
 931                          * checksum, thus giving us the complete TCP checksum.
 932                          *
 933                          * To put it in simple terms, if checksum(1,2,3,4)=10,
 934                          * then checksum(1,2,3,4,5) == checksum(10,5).
 935                          * This property is what allows us to "cheat" and
 936                          * checksum only the IP payload which has the TCP
 937                          * th_sum field populated with the pseudo header's
 938                          * checksum, and not need to futz around checksumming
 939                          * pseudo header bytes and TCP header/data in one hit.
 940                          * Refer to RFC 1071 for more info.
 941                          *
 942                          * NB: in_cksum_skip(struct mbuf *m, int len, int skip)
 943                          * in_cksum_skip 2nd argument is NOT the number of
 944                          * bytes to read from the mbuf at "skip" bytes offset
 945                          * from the start of the mbuf (very counter intuitive!).
 946                          * The number of bytes to read is calculated internally
 947                          * by the function as len-skip i.e. to sum over the IP
 948                          * payload (TCP header + data) bytes, it is INCORRECT
 949                          * to call the function like this:
 950                          * in_cksum_skip(at, ip->ip_len - offset, offset)
 951                          * Rather, it should be called like this:
 952                          * in_cksum_skip(at, ip->ip_len, offset)
 953                          * which means read "ip->ip_len - offset" bytes from
 954                          * the mbuf cluster "at" at offset "offset" bytes from
 955                          * the beginning of the "at" mbuf's data pointer.
 956                          */
 957                         th->th_sum = in_cksum_skip(*m, ntohs(ip->ip_len),
 958                             ip_hl);
 959                 }
 960
 961                 /*
 962                  * XXX: Having to calculate the checksum in software and then
 963                  * hash over all bytes is really inefficient. Would be nice to
 964                  * find a way to create the hash and checksum in the same pass
 965                  * over the bytes.
 966                  */
 967                 pn->hash = hash_pkt(*m, ip_hl);
 968         }
 969
 970         mtx_lock(&siftr_pkt_queue_mtx);
 971         STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
 972         mtx_unlock(&siftr_pkt_queue_mtx);
 973         goto ret;
 974
 975 inp_unlock:
 976         if (inp_locally_locked)
 977                 INP_RUNLOCK(inp);
 978
 979 ret:
 980         /* Returning 0 ensures pfil will not discard the pkt */
 981         return (0);
 982 }
 983
 984
 985 #ifdef SIFTR_IPV6
 986 static int
 987 siftr_chkpkt6(void *arg, struct mbuf **m, struct ifnet *ifp, int dir,
 988     struct inpcb *inp)
 989 {
 990         struct pkt_node *pn;
 991         struct ip6_hdr *ip6;
 992         struct tcphdr *th;
 993         struct tcpcb *tp;
 994         struct siftr_stats *ss;
 995         unsigned int ip6_hl;
 996         int inp_locally_locked;
 997
 998         inp_locally_locked = 0;
 999         ss = DPCPU_PTR(ss);
1000
1001         /*
1002          * m_pullup is not required here because ip6_{input|output}
1003          * already do the heavy lifting for us.
1004          */
1005
1006         ip6 = mtod(*m, struct ip6_hdr *);
1007
1008         /*
1009          * Only continue processing if the packet is TCP
1010          * XXX: We should follow the next header fields
1011          * as shown on Pg 6 RFC 2460, but right now we'll
1012          * only check pkts that have no extension headers.
1013          */
1014         if (ip6->ip6_nxt != IPPROTO_TCP)
1015                 goto ret6;
1016
1017         /*
1018          * If a kernel subsystem reinjects packets into the stack, our pfil
1019          * hook will be called multiple times for the same packet.
1020          * Make sure we only process unique packets.
1021          */
1022         if (siftr_chkreinject(*m, dir, ss))
1023                 goto ret6;
1024
1025         if (dir == PFIL_IN)
1026                 ss->n_in++;
1027         else
1028                 ss->n_out++;
1029
1030         ip6_hl = sizeof(struct ip6_hdr);
1031
1032         /*
1033          * Create a tcphdr struct starting at the correct offset
1034          * in the ipv6 packet. ip->ip_hl gives the ip header length
1035          * in 4-byte words, so multiply it to get the size in bytes.
1036          */
1037         th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl);
1038
1039         /*
1040          * For inbound packets, the pfil hooks don't provide a pointer to the
1041          * inpcb, so we need to find it ourselves and lock it.
1042          */
1043         if (!inp) {
1044                 /* Find the corresponding inpcb for this pkt. */
1045                 inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m,
1046                     th->th_sport, th->th_dport, dir, ss);
1047
1048                 if (inp == NULL)
1049                         goto ret6;
1050                 else
1051                         inp_locally_locked = 1;
1052         }
1053
1054         /* Find the TCP control block that corresponds with this packet. */
1055         tp = intotcpcb(inp);
1056
1057         /*
1058          * If we can't find the TCP control block (happens occasionaly for a
1059          * packet sent during the shutdown phase of a TCP connection),
1060          * or we're in the timewait state, bail.
1061          */
1062         if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) {
1063                 if (dir == PFIL_IN)
1064                         ss->nskip_in_tcpcb++;
1065                 else
1066                         ss->nskip_out_tcpcb++;
1067
1068                 goto inp_unlock6;
1069         }
1070
1071         pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO);
1072
1073         if (pn == NULL) {
1074                 if (dir == PFIL_IN)
1075                         ss->nskip_in_malloc++;
1076                 else
1077                         ss->nskip_out_malloc++;
1078
1079                 goto inp_unlock6;
1080         }
1081
1082         siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked);
1083
1084         /* XXX: Figure out how to generate hashes for IPv6 packets. */
1085
1086         mtx_lock(&siftr_pkt_queue_mtx);
1087         STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes);
1088         mtx_unlock(&siftr_pkt_queue_mtx);
1089         goto ret6;
1090
1091 inp_unlock6:
1092         if (inp_locally_locked)
1093                 INP_RUNLOCK(inp);
1094
1095 ret6:
1096         /* Returning 0 ensures pfil will not discard the pkt. */
1097         return (0);
1098 }
1099 #endif /* #ifdef SIFTR_IPV6 */
1100
1101
1102 static int
1103 siftr_pfil(int action)
1104 {
1105         struct pfil_head *pfh_inet;
1106 #ifdef SIFTR_IPV6
1107         struct pfil_head *pfh_inet6;
1108 #endif
1109         VNET_ITERATOR_DECL(vnet_iter);
1110
1111         VNET_LIST_RLOCK();
1112         VNET_FOREACH(vnet_iter) {
1113                 CURVNET_SET(vnet_iter);
1114                 pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET);
1115 #ifdef SIFTR_IPV6
1116                 pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6);
1117 #endif
1118
1119                 if (action == HOOK) {
1120                         pfil_add_hook(siftr_chkpkt, NULL,
1121                             PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1122 #ifdef SIFTR_IPV6
1123                         pfil_add_hook(siftr_chkpkt6, NULL,
1124                             PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1125 #endif
1126                 } else if (action == UNHOOK) {
1127                         pfil_remove_hook(siftr_chkpkt, NULL,
1128                             PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet);
1129 #ifdef SIFTR_IPV6
1130                         pfil_remove_hook(siftr_chkpkt6, NULL,
1131                             PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6);
1132 #endif
1133                 }
1134                 CURVNET_RESTORE();
1135         }
1136         VNET_LIST_RUNLOCK();
1137
1138         return (0);
1139 }
1140
1141
1142 static int
1143 siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS)
1144 {
1145         struct alq *new_alq;
1146         int error;
1147
1148         if (req->newptr == NULL)
1149                 goto skip;
1150
1151         /* If old filename and new filename are different. */
1152         if (strncmp(siftr_logfile, (char *)req->newptr, PATH_MAX)) {
1153
1154                 error = alq_open(&new_alq, req->newptr, curthread->td_ucred,
1155                     SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1156
1157                 /* Bail if unable to create new alq. */
1158                 if (error)
1159                         return (1);
1160
1161                 /*
1162                  * If disabled, siftr_alq == NULL so we simply close
1163                  * the alq as we've proved it can be opened.
1164                  * If enabled, close the existing alq and switch the old
1165                  * for the new.
1166                  */
1167                 if (siftr_alq == NULL)
1168                         alq_close(new_alq);
1169                 else {
1170                         alq_close(siftr_alq);
1171                         siftr_alq = new_alq;
1172                 }
1173         }
1174
1175 skip:
1176         return (sysctl_handle_string(oidp, arg1, arg2, req));
1177 }
1178
1179
1180 static int
1181 siftr_manage_ops(uint8_t action)
1182 {
1183         struct siftr_stats totalss;
1184         struct timeval tval;
1185         struct flow_hash_node *counter, *tmp_counter;
1186         struct sbuf *s;
1187         int i, key_index, ret, error;
1188         uint32_t bytes_to_write, total_skipped_pkts;
1189         uint16_t lport, fport;
1190         uint8_t *key, ipver;
1191
1192 #ifdef SIFTR_IPV6
1193         uint32_t laddr[4];
1194         uint32_t faddr[4];
1195 #else
1196         uint8_t laddr[4];
1197         uint8_t faddr[4];
1198 #endif
1199
1200         error = 0;
1201         total_skipped_pkts = 0;
1202
1203         /* Init an autosizing sbuf that initially holds 200 chars. */
1204         if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL)
1205                 return (-1);
1206
1207         if (action == SIFTR_ENABLE) {
1208                 /*
1209                  * Create our alq
1210                  * XXX: We should abort if alq_open fails!
1211                  */
1212                 alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred,
1213                     SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0);
1214
1215                 STAILQ_INIT(&pkt_queue);
1216
1217                 DPCPU_ZERO(ss);
1218
1219                 siftr_exit_pkt_manager_thread = 0;
1220
1221                 ret = kthread_add(&siftr_pkt_manager_thread, NULL, NULL,
1222                     &siftr_pkt_manager_thr, RFNOWAIT, 0,
1223                     "siftr_pkt_manager_thr");
1224
1225                 siftr_pfil(HOOK);
1226
1227                 microtime(&tval);
1228
1229                 sbuf_printf(s,
1230                     "enable_time_secs=%jd\tenable_time_usecs=%06ld\t"
1231                     "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t"
1232                     "sysver=%u\tipmode=%u\n",
1233                     (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz,
1234                     TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE);
1235
1236                 sbuf_finish(s);
1237                 alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK);
1238
1239         } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) {
1240                 /*
1241                  * Remove the pfil hook functions. All threads currently in
1242                  * the hook functions are allowed to exit before siftr_pfil()
1243                  * returns.
1244                  */
1245                 siftr_pfil(UNHOOK);
1246
1247                 /* This will block until the pkt manager thread unlocks it. */
1248                 mtx_lock(&siftr_pkt_mgr_mtx);
1249
1250                 /* Tell the pkt manager thread that it should exit now. */
1251                 siftr_exit_pkt_manager_thread = 1;
1252
1253                 /*
1254                  * Wake the pkt_manager thread so it realises that
1255                  * siftr_exit_pkt_manager_thread == 1 and exits gracefully.
1256                  * The wakeup won't be delivered until we unlock
1257                  * siftr_pkt_mgr_mtx so this isn't racy.
1258                  */
1259                 wakeup(&wait_for_pkt);
1260
1261                 /* Wait for the pkt_manager thread to exit. */
1262                 mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT,
1263                     "thrwait", 0);
1264
1265                 siftr_pkt_manager_thr = NULL;
1266                 mtx_unlock(&siftr_pkt_mgr_mtx);
1267
1268                 totalss.n_in = DPCPU_VARSUM(ss, n_in);
1269                 totalss.n_out = DPCPU_VARSUM(ss, n_out);
1270                 totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc);
1271                 totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc);
1272                 totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx);
1273                 totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx);
1274                 totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb);
1275                 totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb);
1276                 totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb);
1277                 totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb);
1278
1279                 total_skipped_pkts = totalss.nskip_in_malloc +
1280                     totalss.nskip_out_malloc + totalss.nskip_in_mtx +
1281                     totalss.nskip_out_mtx + totalss.nskip_in_tcpcb +
1282                     totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb +
1283                     totalss.nskip_out_inpcb;
1284
1285                 microtime(&tval);
1286
1287                 sbuf_printf(s,
1288                     "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t"
1289                     "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t"
1290                     "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t"
1291                     "num_outbound_skipped_pkts_malloc=%u\t"
1292                     "num_inbound_skipped_pkts_mtx=%u\t"
1293                     "num_outbound_skipped_pkts_mtx=%u\t"
1294                     "num_inbound_skipped_pkts_tcpcb=%u\t"
1295                     "num_outbound_skipped_pkts_tcpcb=%u\t"
1296                     "num_inbound_skipped_pkts_inpcb=%u\t"
1297                     "num_outbound_skipped_pkts_inpcb=%u\t"
1298                     "total_skipped_tcp_pkts=%u\tflow_list=",
1299                     (intmax_t)tval.tv_sec,
1300                     tval.tv_usec,
1301                     (uintmax_t)totalss.n_in,
1302                     (uintmax_t)totalss.n_out,
1303                     (uintmax_t)(totalss.n_in + totalss.n_out),
1304                     totalss.nskip_in_malloc,
1305                     totalss.nskip_out_malloc,
1306                     totalss.nskip_in_mtx,
1307                     totalss.nskip_out_mtx,
1308                     totalss.nskip_in_tcpcb,
1309                     totalss.nskip_out_tcpcb,
1310                     totalss.nskip_in_inpcb,
1311                     totalss.nskip_out_inpcb,
1312                     total_skipped_pkts);
1313
1314                 /*
1315                  * Iterate over the flow hash, printing a summary of each
1316                  * flow seen and freeing any malloc'd memory.
1317                  * The hash consists of an array of LISTs (man 3 queue).
1318                  */
1319                 for (i = 0; i <= siftr_hashmask; i++) {
1320                         LIST_FOREACH_SAFE(counter, counter_hash + i, nodes,
1321                             tmp_counter) {
1322                                 key = counter->key;
1323                                 key_index = 1;
1324
1325                                 ipver = key[0];
1326
1327                                 memcpy(laddr, key + key_index, sizeof(laddr));
1328                                 key_index += sizeof(laddr);
1329                                 memcpy(&lport, key + key_index, sizeof(lport));
1330                                 key_index += sizeof(lport);
1331                                 memcpy(faddr, key + key_index, sizeof(faddr));
1332                                 key_index += sizeof(faddr);
1333                                 memcpy(&fport, key + key_index, sizeof(fport));
1334
1335 #ifdef SIFTR_IPV6
1336                                 laddr[3] = ntohl(laddr[3]);
1337                                 faddr[3] = ntohl(faddr[3]);
1338
1339                                 if (ipver == INP_IPV6) {
1340                                         laddr[0] = ntohl(laddr[0]);
1341                                         laddr[1] = ntohl(laddr[1]);
1342                                         laddr[2] = ntohl(laddr[2]);
1343                                         faddr[0] = ntohl(faddr[0]);
1344                                         faddr[1] = ntohl(faddr[1]);
1345                                         faddr[2] = ntohl(faddr[2]);
1346
1347                                         sbuf_printf(s,
1348                                             "%x:%x:%x:%x:%x:%x:%x:%x;%u-"
1349                                             "%x:%x:%x:%x:%x:%x:%x:%x;%u,",
1350                                             UPPER_SHORT(laddr[0]),
1351                                             LOWER_SHORT(laddr[0]),
1352                                             UPPER_SHORT(laddr[1]),
1353                                             LOWER_SHORT(laddr[1]),
1354                                             UPPER_SHORT(laddr[2]),
1355                                             LOWER_SHORT(laddr[2]),
1356                                             UPPER_SHORT(laddr[3]),
1357                                             LOWER_SHORT(laddr[3]),
1358                                             ntohs(lport),
1359                                             UPPER_SHORT(faddr[0]),
1360                                             LOWER_SHORT(faddr[0]),
1361                                             UPPER_SHORT(faddr[1]),
1362                                             LOWER_SHORT(faddr[1]),
1363                                             UPPER_SHORT(faddr[2]),
1364                                             LOWER_SHORT(faddr[2]),
1365                                             UPPER_SHORT(faddr[3]),
1366                                             LOWER_SHORT(faddr[3]),
1367                                             ntohs(fport));
1368                                 } else {
1369                                         laddr[0] = FIRST_OCTET(laddr[3]);
1370                                         laddr[1] = SECOND_OCTET(laddr[3]);
1371                                         laddr[2] = THIRD_OCTET(laddr[3]);
1372                                         laddr[3] = FOURTH_OCTET(laddr[3]);
1373                                         faddr[0] = FIRST_OCTET(faddr[3]);
1374                                         faddr[1] = SECOND_OCTET(faddr[3]);
1375                                         faddr[2] = THIRD_OCTET(faddr[3]);
1376                                         faddr[3] = FOURTH_OCTET(faddr[3]);
1377 #endif
1378                                         sbuf_printf(s,
1379                                             "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,",
1380                                             laddr[0],
1381                                             laddr[1],
1382                                             laddr[2],
1383                                             laddr[3],
1384                                             ntohs(lport),
1385                                             faddr[0],
1386                                             faddr[1],
1387                                             faddr[2],
1388                                             faddr[3],
1389                                             ntohs(fport));
1390 #ifdef SIFTR_IPV6
1391                                 }
1392 #endif
1393
1394                                 free(counter, M_SIFTR_HASHNODE);
1395                         }
1396
1397                         LIST_INIT(counter_hash + i);
1398                 }
1399
1400                 sbuf_printf(s, "\n");
1401                 sbuf_finish(s);
1402
1403                 i = 0;
1404                 do {
1405                         bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i);
1406                         alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK);
1407                         i += bytes_to_write;
1408                 } while (i < sbuf_len(s));
1409
1410                 alq_close(siftr_alq);
1411                 siftr_alq = NULL;
1412         }
1413
1414         sbuf_delete(s);
1415
1416         /*
1417          * XXX: Should be using ret to check if any functions fail
1418          * and set error appropriately
1419          */
1420
1421         return (error);
1422 }
1423
1424
1425 static int
1426 siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS)
1427 {
1428         if (req->newptr == NULL)
1429                 goto skip;
1430
1431         /* If the value passed in isn't 0 or 1, return an error. */
1432         if (CAST_PTR_INT(req->newptr) != 0 && CAST_PTR_INT(req->newptr) != 1)
1433                 return (1);
1434
1435         /* If we are changing state (0 to 1 or 1 to 0). */
1436         if (CAST_PTR_INT(req->newptr) != siftr_enabled )
1437                 if (siftr_manage_ops(CAST_PTR_INT(req->newptr))) {
1438                         siftr_manage_ops(SIFTR_DISABLE);
1439                         return (1);
1440                 }
1441
1442 skip:
1443         return (sysctl_handle_int(oidp, arg1, arg2, req));
1444 }
1445
1446
1447 static void
1448 siftr_shutdown_handler(void *arg)
1449 {
1450         siftr_manage_ops(SIFTR_DISABLE);
1451 }
1452
1453
1454 /*
1455  * Module is being unloaded or machine is shutting down. Take care of cleanup.
1456  */
1457 static int
1458 deinit_siftr(void)
1459 {
1460         /* Cleanup. */
1461         siftr_manage_ops(SIFTR_DISABLE);
1462         hashdestroy(counter_hash, M_SIFTR, siftr_hashmask);
1463         mtx_destroy(&siftr_pkt_queue_mtx);
1464         mtx_destroy(&siftr_pkt_mgr_mtx);
1465
1466         return (0);
1467 }
1468
1469
1470 /*
1471  * Module has just been loaded into the kernel.
1472  */
1473 static int
1474 init_siftr(void)
1475 {
1476         EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL,
1477             SHUTDOWN_PRI_FIRST);
1478
1479         /* Initialise our flow counter hash table. */
1480         counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR,
1481             &siftr_hashmask);
1482
1483         mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF);
1484         mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF);
1485
1486         /* Print message to the user's current terminal. */
1487         uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n"
1488             "          http://caia.swin.edu.au/urp/newtcp\n\n",
1489             MODVERSION_STR);
1490
1491         return (0);
1492 }
1493
1494
1495 /*
1496  * This is the function that is called to load and unload the module.
1497  * When the module is loaded, this function is called once with
1498  * "what" == MOD_LOAD
1499  * When the module is unloaded, this function is called twice with
1500  * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second
1501  * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command,
1502  * this function is called once with "what" = MOD_SHUTDOWN
1503  * When the system is shut down, the handler isn't called until the very end
1504  * of the shutdown sequence i.e. after the disks have been synced.
1505  */
1506 static int
1507 siftr_load_handler(module_t mod, int what, void *arg)
1508 {
1509         int ret;
1510
1511         switch (what) {
1512         case MOD_LOAD:
1513                 ret = init_siftr();
1514                 break;
1515
1516         case MOD_QUIESCE:
1517         case MOD_SHUTDOWN:
1518                 ret = deinit_siftr();
1519                 break;
1520
1521         case MOD_UNLOAD:
1522                 ret = 0;
1523                 break;
1524
1525         default:
1526                 ret = EINVAL;
1527                 break;
1528         }
1529
1530         return (ret);
1531 }
1532
1533
1534 static moduledata_t siftr_mod = {
1535         .name = "siftr",
1536         .evhand = siftr_load_handler,
1537 };
1538
1539 /*
1540  * Param 1: name of the kernel module
1541  * Param 2: moduledata_t struct containing info about the kernel module
1542  *          and the execution entry point for the module
1543  * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h
1544  *          Defines the module initialisation order
1545  * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h
1546  *          Defines the initialisation order of this kld relative to others
1547  *          within the same subsystem as defined by param 3
1548  */
1549 DECLARE_MODULE(siftr, siftr_mod, SI_SUB_SMP, SI_ORDER_ANY);
1550 MODULE_DEPEND(siftr, alq, 1, 1, 1);
1551 MODULE_VERSION(siftr, MODVERSION);